We release the new benchmark Description of the image MMMU for evaluating multi-modal models!